Maximal Onset Principle as a cue for syllable detection at birth



Gonzalo García-Castro

sap2025-mop-newborns | 2025-04-29


gongcastro

Continuous speech, discrete units

From Meyer (2018).

The syllable: a privileged linguistic unit?

Newborns (and adults) preferentially parse the speech signal into syllable-sized units (e.g., Bijeljac-Babic, Bertoncini, and Mehler 1993; Fló et al. 2022; Luo and Poeppel 2007; Jusczyk and Derrah 1987; Bertoncini et al. 1988)

From Fló et al. (2022).

Syllabic structure

Structure Onset Nucleus Coda
V a
CV t a
CVC t a n
VC a n

Language-specific constraints to syllabic structure

Adapted from Özer (2024).
Structure Japanese Spanish English
V u.mi o.jo a.ny
CV ya.ma.ha ca.sa fai.ry
CVC hon.da rin.cón con.trol
CCVC fres.co fresh
CCVCC trans.por.te shrink
CCCVCCC strengths

Maximal Onset Principle (MOP)

Consonants are preferably grouped at syllabic onset

  • MOP+: CV.CCV - /mo.pla/
  • MOP-: CVC.CV - /mop.la/

Are newborns sensitive to (violations of) the MOP?

fNIRS: differential haemodynamic response1 to MOP+ and MOP- stimuli.

Hypothesis 1: If newborns are sensitive to violations of the MOP, the haemodynamic responses to MOP+ and MOP- words should differ (morphology of the signal).

mo.pla \(\neq\) mop.la

Are newborns sensitive to (violations of) the MOP?

Hypothesis 2: If newborns have an innate preference for MOP+ structures, the haemodynamic response to MOP+ words should be greater than for MOP- words (max. amplitude, AUC, time-to-peak). (?)

mo.pla \(\gt\) mop.la

Are newborns sensitive to (violations of) the MOP?

Hypothesis 3: If newborns discriminate between MOP+ and MOP- (Hypothesis 1), they do so by processing the disyllabic structure of the word (not just the first syllable).

mop.pla \(\neq\) mop

Participants

  • Healthy, full term neonates.1
  • Born at the Àrea de Maternitat form Hospital Sant Joan de Déu (Barcelona), tested in their room.
Figure 1: Participant information.

Participants

  • Pilot (2025-02-05–2025-03-06): 13: Participants.
  • Experiment (2025-02-05–now): 50.
    • Excluded (< 10 min.): 18 crying, 3 bad capping.
    • Provided valid data: 31
    • Data quality ranges a lot between participants

Testing setup and procedure

Stimuli

CVCCV words: Onset + Vowel + Consonant cluster (CC) + Vowel

Table 1: Stimuli lists.
List Syllabification Words
List 1 ST (CV-CCV) MO-PLA, SA-KLO, TI-PLE, DE-KLI
DW (CVC-CV) MOP-LA, SAK-LO, TIP-LE, DEK-LI
DS (CVC) MOP, SAK, TIP, DEK
List 2 ST (CV-CCV) MO-KLA, SA-PLO, TI-KLE, DE-PLI
DW (CVC-CV) MOK-LA, SAP-LO, TIK-LE, DEP-LI
DS (CVC) MOK, SAP, TIK, DEP

Stimuli

  • Synthesised using MBROLA (it4 voice) (pymbrola Python package)
  • 25 ms pause between syllables
  • Constant prosody: 200 Hz F0
  • Manually removed offset voicing from CVC syllables (Praat)

Block design

Data analysis

MNE-NIRS (Python):

  1. Light intensity to optical density
  2. Automatic channel rejection based (SCI \(\geq\) .80)
  3. Motion arctifact correction (TDDR)
  4. Calculate \(\Delta\)HbO and \(\Delta\)HbO using the modified Beer-Lambert Law
  5. Band-pass filter (0.01-1.0 Hz)
  6. Block epoching (-5 to 30 seconds), baseline correction, linear detrend
  7. Block averaging
  8. Participant rejection (\(\geq\) 2 trials per condition)

Results

Haemodynamic response averaged across blocks, channels, and participants.

HbO HbR

Results

Haemodynamic response averaged across blocks, channels, and participants.

ST DW DS

Results

Haemodynamic response averaged across blocks and participants.

Results

Haemodynamic response averaged across blocks and participants.

Discussion

So far:

  • Testing protocol done
  • fNIRS setup up and running
  • Data processing pipeline almost ready

Discussion

Next steps:

  • Finish data collection: 30-40 participants.
  • Statistical modelling:
    • Time-domain analysis: Cluster-based permutation testing
    • Waveform analysis: Bayesian GAMMs
    • Summary statistics: Peak amplitude, AUC, time-to-peak

Appendix

NIRS setup

  • NIRSport2 (NIRx), CW 760 nm & 850 nm
  • Sampling frequency 20.345 Hz (~0.05 s samples)
  • NIRScap: 8 channels LH, 8 channels RH

References

Bertoncini, Josiane, Ranka Bijeljac-Babic, Peter W Jusczyk, Lori J Kennedy, and Jacques Mehler. 1988. “An Investigation of Young Infants’ Perceptual Representations of Speech Sounds.” Journal of Experimental Psychology: General 117 (1): 21.
Bijeljac-Babic, Ranka, Josiane Bertoncini, and Jacques Mehler. 1993. “How Do 4-Day-Old Infants Categorize Multisyllabic Utterances?” Developmental Psychology 29 (4): 711.
Fló, Ana, Lucas Benjamin, Marie Palu, and Ghislaine Dehaene-Lambertz. 2022. “Sleeping Neonates Track Transitional Probabilities in Speech but Only Retain the First Syllable of Words.” Scientific Reports 12 (1): 4391.
Jusczyk, Peter W, and Carolyn Derrah. 1987. “Representation of Speech Sounds by Young Infants.” Developmental Psychology 23 (5): 648.
Luo, Huan, and David Poeppel. 2007. “Phase Patterns of Neuronal Responses Reliably Discriminate Speech in Human Auditory Cortex.” Neuron 54 (6): 1001–10.
Meyer, Lars. 2018. “The Neural Oscillations of Speech Processing and Language Comprehension: State of the Art and Emerging Mechanisms.” European Journal of Neuroscience 48 (7): 2609–21.